Localization and path planning represent crucial aspects of autonomous driving systems, due to the complexity and variety of contexts of urban and suburban environments. Prior knowledge of the navigation environment, in the form of highly accurate maps, is therefore often exploited in order to simplify these tasks.

The need for detailed a-priori maps represents one of the major limiting factors for the widespread adoption of autonomous vehicles beyond small areas, for which managing and updating such maps is feasible in terms of resources required.

For this reason, in the last years a growing interest has been put towards developing navigation approaches that are based on topological maps databases, for which many providers are already available. These types of maps are however less accurate and cannot be exploited with classical localization and planning techniques.

In this presentation, we will first discuss an algorithm for large-scale robust vehicle localization based on topological maps. We then exploit this output, together with observations from the on-board perception system, to plan safe and collision-free trajectories for the vehicle. The navigation approach is validated with experimental sessions on public roads open to traffic.